⚠ This notebook has stopped taking notes: newest record is 23 days old (eng-0280), against an expected cadence of 14 days.
clm-0125
measured-heremed ●●○
citable URL: https://halobench.com/records/clm-0125/ — this address never moves; the anchor /records/#clm-0125 keeps resolving
The Qwen3.8-Flash-Next records from 2026-08-27 remain visible as a provisional first look: cfg-0177 and run-0632 record the reasoning-low capability arm, cfg-0178 and run-0633 through run-0636 record the non-served engine-floor depth ladder, and eng-0275 records the joined wall-meter window. They are real observations, not erased failures, but they are not admitted as the current capability leader, energy-per-correct leader, production-throughput result, or role-fit recommendation. The first look did not retain a protocol-valid evidence-bearing screen, formal guard, repeated capability series, exact full-run serving transport, or an unambiguous independently admitted energy denominator. A new intended/default-reasoning run is being prepared to supersede the decision boundary while preserving this history.
Interim current-best-information boundary. The underlying first-look data remain public and citable as historical observations; only their headline admission is held pending protocol-valid superseding evidence.
Cited by — computed at build time, never stored
supersedesclm-0124 — the withdrawn record keeps its URL